KBaba7/llama.cpp

1#include "arg.h"2#include "common.h"3#include "log.h"4#include "llama.h"5 6#include <algorithm>7#include <cstdio>8#include <string>9#include <vector>10 11static void print_usage(int, char ** argv) {12    LOG("\nexample usage:\n");13    LOG("\n    %s -m model.gguf -p \"Hello my name is\" -n 32 -np 4\n", argv[0]);14    LOG("\n");15}16 17int main(int argc, char ** argv) {18    common_params params;19 20    params.prompt = "Hello my name is";21    params.n_predict = 32;22 23    if (!common_params_parse(argc, argv, params, LLAMA_EXAMPLE_COMMON, print_usage)) {24        return 1;25    }26 27    common_init();28 29    // number of parallel batches30    int n_parallel = params.n_parallel;31 32    // total length of the sequences including the prompt33    int n_predict = params.n_predict;34 35    // init LLM36 37    llama_backend_init();38    llama_numa_init(params.numa);39 40    // initialize the model41 42    llama_model_params model_params = common_model_params_to_llama(params);43 44    llama_model * model = llama_model_load_from_file(params.model.c_str(), model_params);45 46    if (model == NULL) {47        LOG_ERR("%s: error: unable to load model\n" , __func__);48        return 1;49    }50 51    const llama_vocab * vocab = llama_model_get_vocab(model);52 53    // tokenize the prompt54 55    std::vector<llama_token> tokens_list;56    tokens_list = common_tokenize(vocab, params.prompt, true);57 58    const int n_kv_req = tokens_list.size() + (n_predict - tokens_list.size())*n_parallel;59 60    // initialize the context61 62    llama_context_params ctx_params = common_context_params_to_llama(params);63 64    ctx_params.n_ctx   = n_kv_req;65    ctx_params.n_batch = std::max(n_predict, n_parallel);66 67    llama_context * ctx = llama_init_from_model(model, ctx_params);68 69    auto sparams = llama_sampler_chain_default_params();70    sparams.no_perf = false;71 72    llama_sampler * smpl = llama_sampler_chain_init(sparams);73 74    llama_sampler_chain_add(smpl, llama_sampler_init_top_k(params.sampling.top_k));75    llama_sampler_chain_add(smpl, llama_sampler_init_top_p(params.sampling.top_p, params.sampling.min_keep));76    llama_sampler_chain_add(smpl, llama_sampler_init_temp (params.sampling.temp));77    llama_sampler_chain_add(smpl, llama_sampler_init_dist (params.sampling.seed));78 79    if (ctx == NULL) {80        LOG_ERR("%s: error: failed to create the llama_context\n" , __func__);81        return 1;82    }83 84    const int n_ctx = llama_n_ctx(ctx);85 86    LOG_INF("\n%s: n_predict = %d, n_ctx = %d, n_batch = %u, n_parallel = %d, n_kv_req = %d\n", __func__, n_predict, n_ctx, ctx_params.n_batch, n_parallel, n_kv_req);87 88    // make sure the KV cache is big enough to hold all the prompt and generated tokens89    if (n_kv_req > n_ctx) {90        LOG_ERR("%s: error: n_kv_req (%d) > n_ctx, the required KV cache size is not big enough\n", __func__,  n_kv_req);91        LOG_ERR("%s:        either reduce n_parallel or increase n_ctx\n", __func__);92        return 1;93    }94 95    // print the prompt token-by-token96 97    LOG("\n");98 99    for (auto id : tokens_list) {100        LOG("%s", common_token_to_piece(ctx, id).c_str());101    }102 103    // create a llama_batch104    // we use this object to submit token data for decoding105    llama_batch batch = llama_batch_init(std::max(tokens_list.size(), (size_t) n_parallel), 0, n_parallel);106 107    std::vector<llama_seq_id> seq_ids(n_parallel, 0);108    for (int32_t i = 0; i < n_parallel; ++i) {109        seq_ids[i] = i;110    }111 112    // evaluate the initial prompt113    for (size_t i = 0; i < tokens_list.size(); ++i) {114        common_batch_add(batch, tokens_list[i], i, seq_ids, false);115    }116    GGML_ASSERT(batch.n_tokens == (int) tokens_list.size());117 118    if (llama_model_has_encoder(model)) {119        if (llama_encode(ctx, batch)) {120            LOG_ERR("%s : failed to eval\n", __func__);121            return 1;122        }123 124        llama_token decoder_start_token_id = llama_model_decoder_start_token(model);125        if (decoder_start_token_id == LLAMA_TOKEN_NULL) {126            decoder_start_token_id = llama_vocab_bos(vocab);127        }128 129        common_batch_clear(batch);130        common_batch_add(batch, decoder_start_token_id, 0, seq_ids, false);131    }132 133    // llama_decode will output logits only for the last token of the prompt134    batch.logits[batch.n_tokens - 1] = true;135 136    if (llama_decode(ctx, batch) != 0) {137        LOG_ERR("%s: llama_decode() failed\n", __func__);138        return 1;139    }140 141    //// assign the system KV cache to all parallel sequences142    //// this way, the parallel sequences will "reuse" the prompt tokens without having to copy them143    //for (int32_t i = 1; i < n_parallel; ++i) {144    //    llama_kv_cache_seq_cp(ctx, 0, i, -1, -1);145    //}146 147    if (n_parallel > 1) {148        LOG("\n\n%s: generating %d sequences ...\n", __func__, n_parallel);149    }150 151    // main loop152 153    // we will store the parallel decoded sequences in this vector154    std::vector<std::string> streams(n_parallel);155 156    // remember the batch index of the last token for each parallel sequence157    // we need this to determine which logits to sample from158    std::vector<int32_t> i_batch(n_parallel, batch.n_tokens - 1);159 160    int n_cur    = batch.n_tokens;161    int n_decode = 0;162 163    const auto t_main_start = ggml_time_us();164 165    while (n_cur <= n_predict) {166        // prepare the next batch167        common_batch_clear(batch);168 169        // sample the next token for each parallel sequence / stream170        for (int32_t i = 0; i < n_parallel; ++i) {171            if (i_batch[i] < 0) {172                // the stream has already finished173                continue;174            }175 176            const llama_token new_token_id = llama_sampler_sample(smpl, ctx, i_batch[i]);177 178            // is it an end of generation? -> mark the stream as finished179            if (llama_vocab_is_eog(vocab, new_token_id) || n_cur == n_predict) {180                i_batch[i] = -1;181                LOG("\n");182                if (n_parallel > 1) {183                    LOG_INF("%s: stream %d finished at n_cur = %d", __func__, i, n_cur);184                }185 186                continue;187            }188 189            // if there is only one stream, we print immediately to stdout190            if (n_parallel == 1) {191                LOG("%s", common_token_to_piece(ctx, new_token_id).c_str());192            }193 194            streams[i] += common_token_to_piece(ctx, new_token_id);195 196            i_batch[i] = batch.n_tokens;197 198            // push this new token for next evaluation199            common_batch_add(batch, new_token_id, n_cur, { i }, true);200 201            n_decode += 1;202        }203 204        // all streams are finished205        if (batch.n_tokens == 0) {206            break;207        }208 209        n_cur += 1;210 211        // evaluate the current batch with the transformer model212        if (llama_decode(ctx, batch)) {213            LOG_ERR("%s : failed to eval, return code %d\n", __func__, 1);214            return 1;215        }216    }217 218    if (n_parallel > 1) {219        LOG("\n");220 221        for (int32_t i = 0; i < n_parallel; ++i) {222            LOG("sequence %d:\n\n%s%s\n\n", i, params.prompt.c_str(), streams[i].c_str());223        }224    }225 226    const auto t_main_end = ggml_time_us();227 228    LOG_INF("%s: decoded %d tokens in %.2f s, speed: %.2f t/s\n",229            __func__, n_decode, (t_main_end - t_main_start) / 1000000.0f, n_decode / ((t_main_end - t_main_start) / 1000000.0f));230 231    LOG("\n");232    llama_perf_sampler_print(smpl);233    llama_perf_context_print(ctx);234 235    fprintf(stderr, "\n");236 237    llama_batch_free(batch);238 239    llama_sampler_free(smpl);240    llama_free(ctx);241    llama_model_free(model);242 243    llama_backend_free();244 245    return 0;246}247